EP2669798A1 - Répartiteur de charge, gestionnaire de ressources intra-grappes, gestionnaire de ressources intergrappe, appareil de traitement de signaux de bande de base, procédé et programme d'ordinateur pour distribuer une charge - Google Patents

Répartiteur de charge, gestionnaire de ressources intra-grappes, gestionnaire de ressources intergrappe, appareil de traitement de signaux de bande de base, procédé et programme d'ordinateur pour distribuer une charge Download PDF

Info

Publication number
EP2669798A1
EP2669798A1 EP12305600.4A EP12305600A EP2669798A1 EP 2669798 A1 EP2669798 A1 EP 2669798A1 EP 12305600 A EP12305600 A EP 12305600A EP 2669798 A1 EP2669798 A1 EP 2669798A1
Authority
EP
European Patent Office
Prior art keywords
load
processing
processors
processor
state
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
EP12305600.4A
Other languages
German (de)
English (en)
Inventor
Fariborz Derakhshan
Peter Schefczik
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Alcatel Lucent SAS
Original Assignee
Alcatel Lucent SAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Alcatel Lucent SAS filed Critical Alcatel Lucent SAS
Priority to EP12305600.4A priority Critical patent/EP2669798A1/fr
Priority to PCT/EP2013/057451 priority patent/WO2013178394A1/fr
Publication of EP2669798A1 publication Critical patent/EP2669798A1/fr
Withdrawn legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00Arrangements for program control, e.g. control units
    • G06F9/06Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46Multiprogramming arrangements
    • G06F9/50Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5005Allocation of resources, e.g. of the central processing unit [CPU] to service a request
    • G06F9/5027Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
    • G06F9/505Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering the load
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02DCLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00Energy efficient computing, e.g. low power processors, power management or thermal management

Definitions

  • Embodiments of the present invention relate to processing systems, more particularly but not exclusively to cloud computing systems in communication systems.
  • BBUs BaseBand processing Units
  • RRUs Remote Radio Units
  • Another application of cloud networks is a data center scenario where, as above, the processing resources within the data centers are aggregated in a virtual pool and dynamically assigned to incoming requests.
  • Embodiments are based on the finding that "cloudification" may follow the virtualization paradigm based on resource pooling and abstraction. In the context of traditional communication networks the allocation of processing units to resource requests is more or less fixed. It is another finding that a better or even optimum dynamic allocation of processing resources to resource demands that arrive in a virtualized heterogeneous resource pool may be desirable. Embodiments may therefore provide a flexible adaptation concept, which can enable a better or even maximum exploitation of processing resources with high allocation efficiency. Embodiments may be versatile and they may allow for application in diverse heterogeneous systems with different kinds of quality measures and policies to control, improve or even optimize dynamic resource selection.
  • Embodiments may provide apparatuses, methods or algorithms to allocate network resources instantaneously to processing requests in an improved or even optimum manner without a-priori knowledge of the arriving request sizes and arrival times.
  • Embodiments may provide an integrated solution that copes with the diametrically opposing strategies of load balancing and load concentration. For each strategy embodiments may dynamically adapt to different network conditions, for example, relaxed, intermediate and stressed load conditions.
  • Embodiments may assign incoming resource requests, i.e. requests for processing resources, to a set of processors such that either the necessary number of processing units is low or even minimized (load concentration) or the processing load on all processing units is balanced (load balancing). Load concentration therefore refers to a strategy where the processing is rather concentrated on a few processing resources such that other resources can be kept idle.
  • Load balancing differs from load concentration in that a use of multiple processing resources is made at low individual load rather than using some of the processing resources at high or maximum capability. Hence, the individual processing tasks can be processed faster.
  • Load balancing is therefore targeting Quality of Service (QoS) in that the processing time per processing request is kept low or at minimum, which is due to a lower average load of the processing resources.
  • QoS Quality of Service
  • Load balancing may help reducing bottlenecks and remote processing, i.e. utilization of processing resources of other clusters due to overload of the local cluster.
  • load balancing may reduce overprovisioning of processing resources with incoming processing requests and may therewith reduce bottlenecks. At the same time it may reduce a need for remote resource allocation, such as inter-cluster allocations, which prolong the overall serving time of a processing request. Load balancing may thereby reduce the average serving time of a processing request.
  • Embodiments provide a load distributor for distributing processing requests in an environment comprising a plurality of processors.
  • embodiments provide an entity, which assigns incoming processing requests to processing capabilities.
  • the processing capabilities can be established by means of a plurality of processors, for example, in a cloud environment.
  • the plurality of processors may be comprised on boards, such as a multiprocessor boards, it may be comprised in multiple processor boards, for example, a plurality of processing boards with a plurality of processors each, or the plurality of processors may correspond to a structure with multiple, potentially distanced computers, each carrying one more processors.
  • the load distributor is operable to assign a processing request to the plurality of processors based on a state of load of the plurality of processors and based on a predetermined distribution criterion.
  • the load distributor may base assignment decisions on processing requests on the load status of the processing capabilities or processors and on a predetermined distribution criterion, which may specify a target load condition or distribution.
  • the predetermined distribution criterion may, for example, correspond to a load balancing criterion or a load concentration criterion.
  • the load distributor may target to even out an average load among the processing capabilities, i.e. among the plurality of processors.
  • the predetermined distribution criterion corresponds to load concentration, then the load distributor may try to concentrate the processing to as few processors as possible, in order to keep as many processors in idle or unused mode as possible. The processors in idle or unused mode may even be switched off, e.g., for energy conservation purposes.
  • CAPital Expenditures (CAPEX) and OPerational EXpenditures (OPEX) may be reduced when using load concentration.
  • the load distributor can be operable to distinguish at least two different states of load of the plurality of processors.
  • the state of load of the processors may correspond to a relative measure determining to what extent the processors capabilities are occupied based on the processors full capacity.
  • the state of load may correspond to an absolute measure, e.g. determining what absolute capacity is occupied, for example, in terms of Mega FLoating-point OPerations (MFLOPs).
  • MFLOPs Mega FLoating-point OPerations
  • the state of load may correspond to a remaining processing capability, for example, MFLOPs a processor has available before reaching its full processing capacity or capability.
  • the load distributor may monitor the processing capabilities of the plurality of processors in order to enable decisions on processing request distributions, which may make efficient use of the plurality of processors in line with the predetermined distribution criterion.
  • the state of load may be determined by the inter-arrival rate of the processing requests.
  • the state of load of the plurality of processors corresponds to a number of processing requests arriving in a predetermined time interval.
  • This state of load variant may also be relative to an average allocation time per processing request to the respective processing resources, which may also be influenced by the individual loads of the processors. That is to say in some embodiments the state of load can be based on a number of arriving processing requests in a predetermined time interval.
  • the state of load in these embodiments is also referred to as occurrence rate of processing requests, which can be absolute or relative.
  • the load distributor may distinguish multiple states of load of the plurality of processors.
  • one of the at least two states may correspond to a relaxed state and another one of the at least two states of load may correspond to a stressed state.
  • the processing load of the plurality of processors is higher in the stressed state than in the relaxed state.
  • the inter-arrival time of processing requests is shorter in the stressed state than in the relaxed state. That is to say the occurrence rate of the processing requests is higher in the stressed state than in the relaxed state.
  • the load distributor may base the processing request distributions on whether the plurality of processors is in stressed mode or in relaxed mode.
  • a load threshold may be predefined in terms of an absolute or a relative load, wherein the plurality of processors is considered in stressed mode, when the load of the processors lies above the threshold and the plurality of processors is considered to be in relaxed mode, when the load of the plurality of processors lies below the threshold.
  • the load of the plurality of processors may be determined by an overall load of all of the plurality of processors, it may be determined in terms of a subset of the plurality of processors, for example, only the processors which are actually in use. Hence, in some embodiments, especially when the predetermined distribution criterion corresponds to load concentration, there may be a subset of processors, which is operated at high loads. Another subset of processors may be switched off or idle. In such embodiments, the load of the plurality of processors may be determined by the number of highly loaded processors compared to the number of unused processors. In other embodiments, the state of load may be determined by the load that is processed by the occupied processors compared to the load the unused processors could process.
  • a third state of load may correspond to an intermediate state, in which the processing load of the plurality of processors is higher than in the relaxed state and lower than in the stressed state.
  • the inter-arrival time of processing requests in the intermediate state is shorter than in the relaxed state and longer than in the stressed state.
  • the occurrence rate of the processing requests is higher than in the relaxed state and lower than in the stressed state. That is to say in some embodiments three states or conditions of the plurality of processing resources or processors are distinguished, a low load condition, which is also referred to as relaxed mode or state, an intermediate state, and a stressed state in which the load or occurrence rate is high.
  • the load distributor can be operable, when in stressed state, to form a sequence of at least a subset of the plurality of processors. That is to say, when in stressed state, the load that is being processed by the plurality of processors may already be high or the occurrence rate of the processing requests is high. Therefore, the load distributor may not spend time on determining the loads of the individual processors and sorting the processors accordingly. Instead, an arbitrary sequence of the plurality of processors may be formed, and the processing request may be assigned to a first processor in the sequence, which provides processing capacity for the processing request. Hence, the load distributor may assign a processing request to any processor of the plurality of processors, for as long as said processor provides the capability to process the request.
  • embodiments may seek to reduce or even minimize the time the load distributor needs to assign an incoming processing request to one or more of the plurality of processors. Embodiments may further try to assign a next incoming processing request to a different processor than the one before. Multiple mechanisms can be used in embodiments to achieve a distribution of the incoming processing requests across the plurality of processors.
  • an incoming processing request may be assigned to the first processor in the sequence, which provides the according capability.
  • Said processor may, after the assignment, be placed at the end of the sequence. That way, if the next incoming processing request occurs, the sequence can be searched from the beginning and the probability that an early processor in the sequence can provide the requested capacity is high.
  • a processing request is assigned to a processor, said processor is moved to the end of the sequence. This way, the first processor in the sequence is the one, for which the longest time has passed since its last assignment occurred and which, statistically, may be most likely to be able to serve the requested capability.
  • the processor may not be moved or repositioned in the sequence at all.
  • the sequence is formed, and an incoming processing request may be assigned to the first processor in the sequence, which provides the capability.
  • the position of the said processor in the sequence, or the position of the subsequent processor thereafter, may be stored.
  • assignment attempts may start with the next processor in the sequence, such that the sequence is scanned in a ring shape manner.
  • the sequence may then correspond to a ring buffer of pointers to the respective processors, where an assignment pointer, i.e. the position in the buffer at which an assignment attempt starts, moves continuously around the ring.
  • the load distributor can be operable to sort at least a subset of the plurality of processors based on the load condition to obtain a sorted sequence of processors.
  • the load distributor may comprise means for sorting, e.g. in terms of a register or memory with addresses or pointers to at least some of the processors.
  • selecting the processors based on their individual loads may be omitted in stressed state.
  • the load distributor can be operable, when in relaxed state and the predetermined load criterion corresponds to load concentration, to assign a processing request to a processor with the highest load in the sequence, which provides processing capacity for the processing request.
  • the load distributor when in relaxed mode, may form a sorted list of at least a subset of the plurality of processors.
  • the processors are sorted in this list or sequence according to the load condition. If the distribution criterion is load concentration, the load distributor may try to load each of the processors up to its individual capacity, before using or switching on another processor. In embodiments, the sequence or list may then be resorted before the next assignment.
  • the load distributor can be operable, when in relaxed state and the predetermined load criterion corresponds to load balancing, to assign a processing request to a processor with the lowest load in the sequence, which provides processing capacity for the processing request.
  • the distribution criterion is load balancing
  • the load distributor may assign an incoming processing request to the processor with the lowest load in the sequence, it may even switch on an unused processor, before assigning two processing requests to the same processor.
  • the load distributor can be operable to subdivide the plurality of processors in at least two partitions of processors.
  • a partition of processors may correspond to a subset or a subgroup of processors from the plurality of processors.
  • the load distributor can be operable, when in intermediate state and the predetermined load criterion corresponds to load concentration, to sort a first partition of processors, to assign a processing request to a processor with the highest load in the first partition, which provides processing capacity for the processing request, and in case the first partition lacks a processor with sufficient capacity, to sort a second partition of processors, and to assign a processing request to a processor with the highest load in the second partition, which provides processing capacity for the processing request.
  • the predetermined load criterion corresponds to load concentration
  • the load distributor may sort only a subset of the plurality of processors. This can be carried out by forming subsets or partitions of processors, which are sorted and searched one by one, in order to arrive at an assignment, i.e. to find a processor, which provides sufficient capability, faster than in relaxed mode.
  • the predetermined distribution criterion corresponds to load concentration
  • the processors of a partition are loaded it as much as possible, before another processor is switched on or assigned a processing request. If the load distributor fails to find an according processor in the first partition, it continues the process with the second partition. If the load distributor manages to assign the incoming processing request to a processor of the first partition, the sorting and searching of the second partition can be omitted, time can be saved.
  • the load distributor can be operable, when in intermediate state and the predetermined load criterion corresponds to load balancing, to sort a first partition of processors, and to assign a processing request to a processor with the lowest load in the first partition, which provides processing capacity for the processing request.
  • the predetermined distribution criterion corresponds to load balancing
  • the load distributor may try to balance the load within a partition. If, however, the first partition cannot provide a processor with a high enough capacity to process the request, the load distributor may continue to seek a processor in the second partition.
  • the load distributor can be operable to sort a second partition of processors, to assign a processing request to a processor with the lowest load in the second partition, which provides capacity for the processing request. In other words, if the load distributor succeeds in the first partition, time and efforts for sorting and seeking in the second partition can be saved.
  • the load distributor may comprise a plurality of processing units, where a processing unit comprises one or more processors, for example, in terms of a board with multiple processors.
  • the load distributor can be operable to distribute the processing requests among the processing units, and the load distributor can be operable, in case the plurality of processing units lacks capacity for processing a processing request, to forward the processing request to a resource manager comprising further processing capacities.
  • the load distributor may not assign the processing requests on a processor level but on a processing unit level.
  • embodiments of the load distributor may be employed to distribute the load among actual individual processors, among processing units, where each processing unit may comprise multiple processors, or even among entities, which comprise multiple processing units. For example, a processing unit with processors may be managed and the load distributor may distribute incoming processing requests among the processors of the processing unit.
  • the load distributor may correspond to an intra-cluster resource manager, which distributes the load or incoming processing requests within the cluster.
  • the load distributor can correspond to an intra-cluster resource manager, managing the resources or the processing requests of multiple processing units in a cluster.
  • the load distributor may correspond to an inter-cluster resource manager, which distributes the load among multiple clusters, each of which may comprise multiple processing units.
  • embodiments also provide an intra-cluster resource manager comprising the above described load distributor.
  • the plurality of processors may then correspond to one or more processing units, a processing unit may comprise one or more processors.
  • the intra-cluster resource manager can be operable to provide information on the load condition of the cluster of processing units to an inter-cluster resource manager.
  • multiple load distributors can be comprised in a hierarchical structure.
  • One load distributor may correspond to a local resource manager, managing and distributing the loads of multiple processors of a processing unit, another load distributor may manage and distribute the load on multiple processing units.
  • the first load distributor also referred to as local resource manager, may therefore report information on the load condition on the processors within its processing unit to the load distributor above, which is also referred to as intra-cluster resource manager.
  • the information on the load condition may correspond to the individual load conditions of the individual processors. Since this may be, especially when the number of processors is high, a large amount of information, average values may be provided as information on the load condition.
  • the information on the load condition of a cluster may comprise information on an envelope of the load state characteristic in the cluster.
  • Embodiments may also provide an inter-cluster resource manager comprising the load distributor according to the above description.
  • the inter-cluster resource manager can be operable to distribute the processing requests among a plurality of intra-cluster resource managers comprising the plurality of processors or processing units.
  • the above-described intra-cluster resource manager may report information on the load condition in its cluster to an inter-cluster resource manager, which may distribute incoming processing requests among multiple clusters.
  • a baseband processing unit may comprise an embodiment of the load distributor.
  • Embodiments may provide an apparatus for processing baseband signals of a base station transceiver of a mobile communication system, comprising a plurality of processors and the load distributor according to the above description.
  • the RAN architecture may comprise a base station transceiver, which is composed of a Remote Radio Head (RRH) or RRU, which establishes a radio front end, i.e. which receives and transmits the actual radio signals and establishes at least one radio cell.
  • RRHs may provide Transmission (Tx), Reception (Rx) and antenna functions and they can be spatially separated from associated BBUs so that the latter can be combined within a cluster of several BBUs.
  • the mobile communication system may, for example, correspond to one of the 3 rd Generation Partnership Program (3GPP)-standardized mobile communication networks, as e.g.
  • 3GPP 3 rd Generation Partnership Program
  • LTE Long Term Evolution
  • LTE-A LTE-Advanced
  • UMTS Universal Mobile Telecommunication System
  • UTRAN Universal Mobile Telecommunication System
  • E-UTRAN Evolved-UTRAN
  • GSM Global System for Mobile Communication
  • EDGE Enhanced Data Rates for GSM Evolution
  • GERAN GSM/EDGE Radio Access Network
  • OFDMA Orthogonal Frequency Division Multiple Access
  • WI-MAX Worldwide Interoperability for Microwave Access
  • Embodiments may also provide a method for distributing processing requests in an environment comprising a plurality of processors.
  • the method comprises assigning a processing request to the plurality of processors based on a state of load of the plurality of processors and based on a predetermined distribution criterion.
  • Embodiments may also provide a computer program having a program code for performing one of the above-described methods, when the computer program is executed on a computer or processor.
  • Embodiments may provide a flexible adaptation concept that can enable a maximum exploitation of processing resources with high allocation efficiency. This can be achieved through adaptive dynamic resource allocation mechanisms that dynamically adapt to the network state by considering the incoming resource requests and monitoring the available state of resources.
  • the resource allocation strategies applied in the traditional networks of today with rigid interconnectivity may not be able to cope with the new networking paradigms based on virtualized connectivity.
  • the flexible allocation of radio resource processing to available processing resources may not be possible due to the fixed association of antennas to base stations. In the future with flexible baseband pools this situation may change.
  • Embodiments may allow for application in diverse heterogeneous systems with different kinds of quality measures and policies to control and optimize the dynamic resource selection.
  • Embodiments may solve online resource allocation problems in distributed systems like cloud environments comprising mobile and fixed networks.
  • Embodiments may take into account the local processing resources and remotely connected processing resources and they may be tuned to incorporate energy savings and policy rulings. Dynamic calculation of the best resources for a given request may be enabled. Further cost and energy advantages can be achieved when the total number of necessary boards to operate the communication network, e.g. the RAN, can be further reduced.
  • Embodiments may provide optimal or near-optimal resources to execute network requests by adapting the allocation method according to optimization goals and the resource state of the system.
  • Embodiments may further be scalable concerning the number of boards or processors monitored as well as the number of requests processed in parallel.
  • Embodiments may handle variations of load, link quality or other QoS parameters (processing load, delay, etc.), and failures in an improved or even optimal and policy-based way, and may provide optimal resource usage and permit optimal data transport between concerned boards.
  • QoS parameters processing load, delay, etc.
  • a balance of processing load and also a balance of measurement overhead can be achieved and the proper usage of the scarce resources may even be guaranteed.
  • Some embodiments comprise a digital control circuit installed within an apparatus for performing a method.
  • a digital control circuit e.g. a digital signal processor (DSP)
  • DSP digital signal processor
  • yet further embodiments also provide a computer program having a program code for performing embodiments of the method, when the computer program is executed on a computer or a digital processor.
  • Fig. 1 illustrates an embodiment of a load distributor 10 for distributing processing requests in an environment comprising a plurality of processors 20.
  • the load distributor 10 is operable to assign a processing request to the plurality of processors 20 based on a state of load of the plurality of processors 20 and based on a predetermined distribution criterion.
  • the plurality of processors 20 is shown in the embodiment of Fig. 1 as a number of single processors 20, the plurality of processors 20 may comprise further components or may be structured differently as the following embodiments will show.
  • Fig. 2 illustrates embodiments in a hierarchical structure. Fig. 2 shows a number of Processing Clusters PC1, PC2, ..., PCq.
  • Each of the processing clusters comprises one or more Processing Units (PU 11 , PU 12 , PU 1p , PU 21 , PU 22 , PU 2p , PU q1 , PU q2 , PU qp ).
  • each of the PCs in Fig. 2 comprises the same number p of PUs, in other embodiments different PCs may comprise a different number of PUs, correspondingly, each PU may comprise a different number of processors, which can also vary in their type.
  • Fig. 2 shows that each of the PUs comprises a Local Resource Manager (LRM), which is indicated by circular symbols.
  • LRM Local Resource Manager
  • PUs of a PC are managed by an Intra-Cluster Resource Manager (IntraClusterRM), which is indicated in Fig. 2 by triangular symbols.
  • Fig. 2 shows Inter-Cluster Resource Managers (InterClusterRM), which are represented by diamond shaped symbols.
  • Each of the hierarchical administration components may comprise a load distributor 10 in line with Fig. 1 . That is to say, the InterClusterRM, the IntraClusterRM, and/or the LRM of Fig. 2 may comprise a load distributor according to Fig. 1 .
  • the IntraClusterRM can comprise a load distributor
  • the plurality of processors 20 may correspond to a cluster of PUs
  • a PU may comprise one or more processors.
  • the InterClusterRM may also comprise a load distributor.
  • the InterClusterRM can be operable to distribute the processing requests among a plurality of IntraClusterRMs comprising the plurality of processors.
  • the resource requests are first processed by the IntraClusterRMs, at which they arrive.
  • An IntraClusterRM controls the internal resources of an PC comprising a multiplicity of PUs, where each PU contains one or multiple processors.
  • Such a cluster can, for example, correspond to a virtual MultiSite MultiStandard BaseBand Unit (MSS-BBU) in a cloud-RAN or a data center or any similar virtual pool of resources.
  • MSS-BBU virtual MultiSite MultiStandard BaseBand Unit
  • Some embodiments are applied in a mobile cloud radio access network scenario.
  • the decoupling of the BBUs from the RRUs may allow for collecting the BBU resources in a virtual baseband processing pool and flexibly assign them according to the arriving requests.
  • an embodiment of an apparatus for processing baseband signals of a base station transceiver of a mobile communication system may accordingly comprise a plurality of processors 20 and the above described load distributor 10.
  • Another application of an embodiment may be a data center scenario where the processing resources within the data centers are aggregated in a virtual pool and dynamically assigned to incoming requests.
  • the LRMs which each comprise a load distributor 10, monitor the resources of the processors on each PU and report the current resource state per push implementation to the IntraClusterRMs. If the available internal resources of a processing cluster are not sufficient to meet a resource request, the Intra-ClusterRM sends the request to the collocated InterClusterRM for remote processing. The InterClusterRM then sends the request to its peer InterClusterRMs on other PCs it interacts with.
  • the predetermined distribution criterion corresponds to a load balancing criterion or a load concentration criterion.
  • Other optimization goals like minimization of the overall serving time for requests, or energy optimal strategies can also be taken into account in other embodiments.
  • An example in an embodiment would be a service request of 1 Mbps DL traffic at a Channel Quality Indicator (CQI) value of 15 for a wireless transmission link in a cell of an LTE or LTE-A system.
  • CQI Channel Quality Indicator
  • the resources required to fulfill the request depend on the current state of the connection denoted by the CQI value reported by User Equipment (UE) as, for example, a mobile transceiver.
  • UE User Equipment
  • the resources required may also depend on the UE capability, e.g. if the UE category does only support up to 384 kbps, then less resources will be needed in the cell as compared to when the UE supports 14,4 Mbps.
  • the needed resources can depend on the technique used, like Multiple-Input-Multiple-Output (MIMO), CoOperative MultiPoint (COMP), etc. This can lead to the allocation of different processing resources in the wireless access network, like general purpose Central Processing Units (CPUs), Digital Signal Processors (DSPs), Field Programmable Gate Arrays (FPGAs), Application Specific Integrated Circuits (ASICs), etc. and related links between these elements.
  • MIMO Multiple-Input-Multiple-Output
  • COP CoOperative MultiPoint
  • CPUs Central Processing Units
  • DSPs Digital Signal Processors
  • FPGAs Field Programmable Gate Arrays
  • ASICs Application Specific Integrated Circuits
  • Fig. 3 illustrates a processing capacity view chart illustrating the maximum capacity 300, a tolerable load level 305 and the load margin 310 of a processor. If the load on a processor exceeds the tolerable load level 305, the IntraClusterRM assigns a new coming request either to other processors, or if no processor in the PC can afford the request, it triggers the InterClusterRM to start a remote inter-cluster resource allocation procedure.
  • the network state or load is measured in the following embodiment by the average inter-arrival time of requests compared to the average processing time of the requests.
  • the load distributor 10 is operable to distinguish at least two different states of load of the plurality of processors 20.
  • One of the at least two states corresponds to a relaxed state and another one of the at least two states of load corresponds to a stressed state.
  • the processing load of the plurality of processors is higher in the stressed state than in the relaxed state.
  • the occurrence rate of the processing requests is higher in the stressed state than in the relaxed state.
  • the average inter arrival time of the processing requests is shorter in the stressed state than in the relaxed state.
  • a third state of load corresponds to an intermediate state, in which the processing load of the plurality of processors 20 is higher than in the relaxed state and lower than in the stressed state.
  • the occurrence rate of the processing requests in the intermediate state is higher than in the relaxed state and lower than in the stressed state.
  • the average inter arrival time of the processing requests is shorter in the intermediate state than in the relaxed state and longer than in the stressed state.
  • the resource allocation efficiency is increased by applying the cloud paradigms on the algorithmic design. That is, the rigidity of the resource allocation is broken up and it is dynamically adapted to the current state of the system.
  • Two opposite resource allocation targets are differentiated, which can form the basis of embodiments with adaptive and dynamic resource allocation mechanisms.
  • the first one is load balancing, where the goal is to distribute the processing load uniformly on all available processing units in order to avoid local load peaks and bottlenecks on individual processing units. This increases the availability and scalability of the system and prevents bottlenecks.
  • the second one is load concentration, where the goal is to concentrate the processing load on as few as possible processing units in order to reduce power consumption and inter-PU communication.
  • Embodiments make use of online resource allocation policies like Next Fit, Best Fit, and Worst Fit if necessary.
  • Embodiments using an adaptive dynamic resource allocation algorithm may exploit the complementary advantages of the algorithms by dynamically switching between them on basis of request and resource monitoring.
  • embodiments may use variations of the Worst Fit algorithm and the Next Fit algorithm.
  • the Worst Fit algorithm selects the largest possible free capacity that the request can be assigned to.
  • the Next Fit is a First Fit algorithm with a revolving pointer. Like the First Fit algorithm it starts at the beginning of a processor array or sequence and goes along the array until the first processor that can satisfy the request is found. Then the search pointer is advanced to the next array element after this processor. Thus the next allocation begins searching from where the previous cycle finished.
  • the rationale of this mechanism is to avoid the accumulation of small fragments at the beginning of the processor array.
  • the Best Fit algorithm selects the smallest free capacity that can afford the request.
  • All allocation procedures are based on the same common start phase, in which a service request S arrives at the IntraClusterRM, cf. Fig. 2 , comprising the load distributor 10.
  • the IntraClusterRM retrieves the resource state on each processor on each intra-cluster PU.
  • the IntraClusterRM assigns the requests to the appropriate intra-cluster resources based on the resource allocation targets described above.
  • the first target or predetermined distribution criterion is load balancing.
  • the aim of load balancing is the homogeneous distribution of load on all available processors. It tries to keep the load on all processors, as far as possible, at approximately the same level.
  • the load distributor 10 is operable to sort at least a subset of the plurality of processors 20 based on their load condition to obtain a sorted sequence of processors.
  • the load distributor 10 is further operable, when in relaxed state and the predetermined load criterion corresponds to load balancing, to assign a processing request to a processor with the lowest load in the sequence, which provides processing capacity for the processing request. That is to say, the IntraClusterRM cares about the optimization of the resource efficiency without considering the resource allocation speed. It has enough time to scan the entire processor list.
  • Fig. 4 illustrates an embodiment with load balancing in relaxed state.
  • Fig. 4 shows three sequences of sorted processors 400, 405, and 410, corresponding to three subsequent steps as carried out by the load distributor 10 in the present embodiment.
  • processors of the plurality of processors 20 are numbered individually as 1, 2, 3, 4, and 5.
  • the IntraClusterRM after having received the processing request r1
  • the array is sorted again, resulting in the updated sequence 405.
  • the processor 1, which was assigned the first processing request 1, now has a changed load or capacity, which is indicated as processor 1' in Fig. 4 .
  • the processor 1' is on the third place from the beginning, i.e. now has the third lowest load state.
  • the assignment procedure of the second processing request r2 then starts on the updated sequence 410, as indicated at the bottom of Fig. 4 .
  • the load distributor When in intermediate condition or state, the load distributor is operable to subdivide the plurality of processors 20, comprising individual processors 1, 2, 3, 4, 5, in at least two partitions of processors 20a and 20b, as it is shown in Fig. 5 , which illustrates an embodiment with load balancing in intermediate state.
  • the IntraClusterRM comprising the load distributor 10 takes care of the resource efficiency as well as of the allocation speed. Therefore it divides the processor list 20 into partitions 20a, 20b and scans only a partition 20a of the processor list 20 for sorting.
  • Fig. 5 shows three sequences 500, 505, 510 of partitioned processors, illustrating the process.
  • Sequence 505 shows the situation after the first assignment.
  • the processor with the minimum load in partition 20a is now processor 2 and processor 1' has the highest load. If a second processing request 2 is received, the above procedure starts over again, this time based on the sequence 510, as indicated at the bottom of Fig. 5 .
  • Fig. 6 illustrates an embodiment with load balancing in stressed state or stress condition.
  • Fig. 6 shows three sequences 600, 605, 610 of the plurality of processors 20, which are individually numbered 1, 2, 3, 4, 5.
  • the load distributor 10 is operable, when in stressed state, to form a sequence 600 of at least a subset of the plurality of processors 20, to assign a processing request r1 to a first processor 1 in the sequence 600, which provides processing capacity for the processing request r1.
  • the processor is repositioned in the sequence 605.
  • processor 1 is replaced to the end of the sequence 605.
  • the IntraClusterRM cares about the maximization of the allocation speed without considering the resource efficiency. It has no time for sorting.
  • the IntraClusterRM cares about the maximization of the allocation speed without considering the resource efficiency. It has no time for sorting.
  • This algorithm is based on the Next Fit mechanism, which is fast but suboptimal in resource efficiency. For example, a small resource request can be assigned to a processor with minimum load. Then this processor is shifted to the end of the processor array and may wait in an almost idle state until it again receives a new a job.
  • the other drawback of this mechanism is that due to lack of time it is not possible to compare the resource states of different processors with each other to make a decision for balancing their loads.
  • a prerequisite of any load balancing strategy may be the scanning of the loads of the processors that take part in the load balancing process, for which in the present embodiment, there is no time given the stressed condition.
  • the load distributor 10 may, instead of repositioning the processor 1' to the end of the sequence 605, begin a later assignment r2 of another processing resource subsequent to the first processor in the sequence 600. That is to say, alternatively, the load distributor may start the assignment procedure of processing request r2 using the sequence 600 at the top of Fig. 6 , but starting with processor 2, i.e. using the sequence 600 in a ring-like manner.
  • Fig. 7 illustrates an embodiment with load concentration in relaxed state.
  • the load distributor 10 is operable to sort at least a subset of the plurality of processors 20 based on their load condition to obtain a sorted sequence 700 of processors. Note that Fig. 7 shows the sequence of processors 20, wherein processor 5 has the minimum load.
  • the load distributor 10 is operable, when in relaxed state and the predetermined load criterion corresponds to load concentration, to assign a processing request r1 to a processor 3 with the highest load in the sequence, which provides processing capacity for the processing request r1.
  • the IntraClusterRM cares about the optimization of the resource efficiency without considering the resource allocation speed. It has enough time to scan the entire processor list 20.
  • the IntraClusterRM cares about the optimization of the resource efficiency without considering the resource allocation speed. It has enough time to scan the entire processor list 20.
  • This Best Fit algorithm is resource efficient but may lead to degradation of performance speed due to capacity fragmentation (small capacity holes) at the beginning of the processor array, which reduces the probability of successful resource allocation at first assignment.
  • capacity fragmentation small capacity holes
  • the sequence can be resorted resulting in sequence 705 in Fig. 7 , based on which the next assignment process can be carried out, cf. 710 in Fig. 7 .
  • Fig. 8 illustrates an embodiment with load concentration in intermediate state.
  • the load distributor 10 in the IntraClusterRM is operable to subdivide the plurality of processors 20 in at least two partitions 20a, 20b of processors 1, 2, 3 and 4, 5.
  • the load distributor 10 is further operable, when in intermediate state and the predetermined load criterion corresponds to load concentration, to sort a first partition 20a of processors 1, 2, 3, to assign a processing request r1 to a processor 2 with the highest load in the first partition 20a, which provides processing capacity for the processing request r1.
  • the load distributor 10 is further operable to sort a second partition 20b of processors 4, 5, to assign the processing request r1 to a processor with the highest load in the second partition 20b, which provides processing capacity for the processing request r1.
  • Fig. 8 illustrates in sequence 805 that processor 2', having been assigned the first request r1, keeps its load based position among the processors of partition 20a.
  • Sequence 810 illustrates the assignment of a subsequent second processing request r2, when the above step V. applies, i.e. it is assumed that none of the processors of the first partition 20a can afford the second request r2 and the load distributer 10 checks the second partition 20b for assigning the second request r2.
  • Fig. 9 illustrates an embodiment with load concentration in stressed state.
  • load distributor 10 is operable, to form a sequence 900 of at least a subset of the plurality of processors 20, to assign a processing request r1 to a first processor 1 in the sequence 900, which provides processing capacity for the processing request. Subsequently the processor 1 can be replaced to the end of the sequence 905, or a later assignment of another processing resource r2 may begin subsequent to the first processor 1 in the sequence 900.
  • the IntraClusterRM cares about the maximization of the allocation speed without considering the resource efficiency. It has no time for scanning the array 20 and sorting the loads.
  • the IntraClusterRM cares about the maximization of the allocation speed without considering the resource efficiency. It has no time for scanning the array 20 and sorting the loads.
  • the IntraClusterRM cares about the maximization of the allocation speed without considering the resource efficiency. It has no time for scanning the array 20 and sorting the loads.
  • the IntraClusterRM cares about the maximization of the
  • This algorithm is based on the Next Fit mechanism, which is fast but suboptimal in resource efficiency. For example, a small resource request can be assigned to a processor 1 with minimum load. Then this processor 1' is shifted to the end of the processor array 905 and may wait in an almost idle state until it again receives a new a job.
  • the Next Fit mechanism may also be applied using a ring-like structure of the sequence as explained above.
  • Fig. 10 shows a block diagram of a flow chart of an embodiment of a method for distributing processing requests.
  • the method distributes processing requests in an environment comprising a plurality of processors 20 and comprises assigning 30 a processing request to the plurality of processors 20 based on a state of load of the plurality of processors 20 and based on the predetermined distribution criterion.
  • Fig. 11 shows a block diagram of a flow chart of another embodiment of a method for distributing processing requests.
  • the embodiment of Fig. 11 shows a flow chart for dynamic adaptation.
  • a processing request occurs.
  • step 42 it is checked whether the predetermined distribution criterion corresponds to load balance or load concentration.
  • the case of load balance is described first.
  • the predetermined distribution criterion is load balance the request inter-arrival times are checked in step 44 in order to determine the state of load of the plurality of processors. According to the above description, if the inter-arrival time is long, relaxed state is detected and a Worst Fit assignment procedure 46 is carried out, cf. Fig. 4 .
  • inter-arrival time is medium, intermediate state is detected and a Worst Fit and partitioning assignment procedure 48 is carried out, cf. Fig. 5 . If the inter-arrival time is short, stressed state is detected and a Next Fit assignment procedure 50 is carried out, cf. Fig. 6 .
  • step 42 load concentration is determined as predetermined distribution criterion, the state of load is detected in subsequent step 52 based on the inter-arrival time of the processing requests.
  • the inter-arrival time is long, relaxed state is detected and a Best Fit assignment procedure 54 is carried out, cf. Fig. 7 .
  • a Best Fit and partitioning assignment procedure 56 is carried out, cf. Fig. 8 .
  • the inter-arrival time is short, stressed state is detected and a Next Fit assignment procedure 58 is carried out, cf. Fig. 9 .
  • the resource allocation concepts as introduced for the embodiments above may have the complementary property that they are rather resource efficient or rather fast. Therefore the decision between these two types may depend on the network condition, distinguished in the present embodiment by the three states, relaxed, intermediate and stressed.
  • Resource efficient algorithms that are based on sorting have a time complexity of O( n log n). This effort does not become large for a small pool with few processing units, like in the MSS-BBUs, however it may vastly degrade the allocation efficiency in large pools with numerous processing units, like in data centers. Therefore in stress phases the sorting can be skipped to avoid the growth of allocation queues with the consequence of suboptimum resource efficiency.
  • Self-balancing Binary Search Trees (BST) with a complexity of O(log n ) - as far as no multiple values (processor identities) have the same key (processor load) - might be implemented to enhance the sorting procedure.
  • the load distributor 10 may comprise a plurality of processing units and a processing unit may comprise one or more processors.
  • the load distributor 10 can be operable to distribute the processing requests among the processing units.
  • the load distributor 10 can be operable, in case the plurality of processing units lacks capacity for processing a processing request, to forward the processing request to a resource manager comprising further processing capacities.
  • each processor or processing unit may comprise a set of different processor types.
  • the LRM of each processor or processing unit may push the resource state or information thereon on each processor to the IntraClusterRM. If a request r k comprises multiple requests concerning diverse layers, then the IntraClusterRM assigns each of the multiple requests to the appropriate processor according to the allocation algorithms introduced above.
  • Fig. 12 illustrates a quantization and sampling example of a state of load in an embodiment.
  • Fig. 12 shows a view chart depicting the state of load or resource consumption of a processor or processing unit versus time.
  • Fig. 12 further illustrates the variances of the load with their peak and envelope values. The envelope lies higher than the peak value, e.g. the load margin may lie in between.
  • the intra-cluster resource manager can be operable to provide information on a load condition of the cluster of processing units to an inter-cluster resource manager and the information on the load condition of the cluster comprises information on an envelope of a load state characteristic in the cluster.
  • the IntraClusterRM may report only the envelop of a current resource state of the PU with maximum available resources to an InterClusterRM.
  • the envelop offset can be dynamically computed via monitoring of the state variances.
  • This technique can be used in embodiments to report the state of load between any of the entities, e.g. from processors to LRMs, from LRM to IntraClauster RM, from IntraClusterRM to InterClusterRM, etc.
  • time bands can be used to determine the sampling periods of the envelope values.
  • Some embodiments may be used in wireless access networks.
  • the resources can comprise different CPU resource types on different boards organized in racks or other clusters.
  • the requests may stem from users that need different communication connectivities for their handsets while they move through the mobile network organized in so called cells. Thereby the processing for cells with high load and for cells with low load can be balanced and the resource efficiency as well as the computation speed can be increased.
  • embodiments may be configured for different goals, e.g. energy saving, blocking probability reduction, precedence, reliability, throughput or any combination thereof.
  • an Input/Output (IO) resource may be precious in future cloud based applications.
  • IO Input/Output
  • embodiments may take the cost of inter-process communication into account; these may be combined with additional transport costs.
  • measurements and configuration parameters embodiments may decide in runtime, which resource allocation mechanisms are the most appropriate and should be applied.
  • the measurements can either be taken from the boards in use or by query from the single Intra-ClusterRM of the boards under question. Also there can be a RM internal storage that has stored such data from former measurements or situations.
  • the RM may also collect the needed information and may also carry out quantizing and sampling of the values.
  • the processing resources or processors are comprised within a data center; they may be merged together in a pool and may be dynamically assigned to incoming requests. Constraints may apply on delay requirements, thus including transport delays towards and from the resource utilized, plus processing delays on the resource itself. Inside a data center a structure similar to Figs. 1 and 2 can be assumed. The interconnection of clusters may be different, e.g. in a fat tree to copy with connection blocking and resilience issues.
  • Embodiments may process dynamically incoming service requests arising in a communication network, e.g. a RAN, and may map same to a set of processing requests before allocating to a set of interconnected resources.
  • a communication network e.g. a RAN
  • Embodiments may enable dynamic allocation of processing tasks to available and monitored processing resources. Algorithm and calculation of the resources happen in the distributed RMs (InterCluster-, IntraClusterRMs and RM clients) for a set of processing boards, whereby one or more of these RMs can take the role of a Master RM that can calculate the current best processing place for each newly incoming task, i.e. assume the role of an embodied load distributor 10.
  • the algorithm may take into account the processing resources under load, the resiliency and load of remotely connected resources and may be tuned to incorporate energy savings and policy rulings.
  • embodiments may allow for dynamically calculating the best resource for a given request. Furthermore, cost and energy advantages can be gained when embodiments diminishes the total number of currently used boards and thus of totally needed boards to operate the communication network, e.g. the RAN.
  • Embodiments may even provide optimal resources to execute network requests by changing the allocation method according to optimization goals, current load and state of the system.
  • Embodiments may be scalable concerning the number of boards monitored as well as to the number of requests processed in parallel. Load changes, link quality changes or other changes in QoS parameters (processing load, delay, ...), as well as failures, may be handled in an optimal and policy based way by embodiments to provide optimal resource usage and permit optimal data transport between concerned boards.
  • a balance of processing load and also a balance of measurements overhead can be achieved and the proper usage of the scarce resources may even be guaranteed.
  • Functional blocks denoted as "means for " shall be understood as functional blocks comprising circuitry that is operable for performing a certain function, respectively.
  • a “means for s.th.” may as well be understood as a “means being operable or suited for s.th.”.
  • a means being operable for performing a certain function does, hence, not imply that such means necessarily is performing said function (at a given time instant).
  • any functional blocks labeled as "means”, “means for sorting” may be provided through the use of dedicated hardware, as e.g. a processor or sorter, as well as hardware capable of executing software in association with appropriate software.
  • dedicated hardware e.g. a processor or sorter
  • the functions may be provided by a single dedicated processor, by a single shared processor, or by a plurality of individual processors, some of which may be shared.
  • processor or “controller” should not be construed to refer exclusively to hardware capable of executing software, and may implicitly include, without limitation, digital signal processor (DSP) hardware, network processor, application specific integrated circuit (ASIC), field programmable gate array (FPGA), read only memory (ROM) for storing software, random access memory (RAM), and non-volatile storage.
  • DSP digital signal processor
  • ASIC application specific integrated circuit
  • FPGA field programmable gate array
  • ROM read only memory
  • RAM random access memory
  • non-volatile storage Other hardware, conventional and/or custom, may also be included.
  • any block diagrams herein represent conceptual views of illustrative circuitry embodying the principles of the invention.
  • any flow charts, flow diagrams, state transition diagrams, pseudo code, and the like represent various processes which may be substantially represented in computer readable medium and so executed by a computer or processor, whether or not such computer or processor is explicitly shown.
  • each claim may stand on its own as a separate embodiment. While each claim may stand on its own as a separate embodiment, it is to be noted that - although a dependent claim may refer in the claims to a specific combination with one or more other claims - other embodiments may also include a combination of the dependent claim with the subject matter of each other dependent claim. Such combinations are proposed herein unless it is stated that a specific combination is not intended. Furthermore, it is intended to include also features of a claim to any other independent claim even if this claim is not directly made dependent to the independent claim.
  • a single step may include or may be broken into multiple sub steps. Such sub steps may be included and part of the disclosure of this single step unless explicitly excluded.

Landscapes

  • Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)
EP12305600.4A 2012-05-31 2012-05-31 Répartiteur de charge, gestionnaire de ressources intra-grappes, gestionnaire de ressources intergrappe, appareil de traitement de signaux de bande de base, procédé et programme d'ordinateur pour distribuer une charge Withdrawn EP2669798A1 (fr)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP12305600.4A EP2669798A1 (fr) 2012-05-31 2012-05-31 Répartiteur de charge, gestionnaire de ressources intra-grappes, gestionnaire de ressources intergrappe, appareil de traitement de signaux de bande de base, procédé et programme d'ordinateur pour distribuer une charge
PCT/EP2013/057451 WO2013178394A1 (fr) 2012-05-31 2013-04-10 Répartiteur de charge, gestionnaire de ressources dans le groupe, gestionnaire de ressources entre groupes, appareil de traitement de signaux de bande de base, procédé et programme informatique de répartition de charge

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
EP12305600.4A EP2669798A1 (fr) 2012-05-31 2012-05-31 Répartiteur de charge, gestionnaire de ressources intra-grappes, gestionnaire de ressources intergrappe, appareil de traitement de signaux de bande de base, procédé et programme d'ordinateur pour distribuer une charge

Publications (1)

Publication Number Publication Date
EP2669798A1 true EP2669798A1 (fr) 2013-12-04

Family

ID=48050737

Family Applications (1)

Application Number Title Priority Date Filing Date
EP12305600.4A Withdrawn EP2669798A1 (fr) 2012-05-31 2012-05-31 Répartiteur de charge, gestionnaire de ressources intra-grappes, gestionnaire de ressources intergrappe, appareil de traitement de signaux de bande de base, procédé et programme d'ordinateur pour distribuer une charge

Country Status (2)

Country Link
EP (1) EP2669798A1 (fr)
WO (1) WO2013178394A1 (fr)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108710540A (zh) * 2018-05-23 2018-10-26 深信服网络科技(深圳)有限公司 一种分布式集群中的资源调度方法、装置及设备
US10159012B2 (en) 2011-11-14 2018-12-18 Alcatel Lucent Baseband signal processing cluster
US10601903B2 (en) 2018-05-17 2020-03-24 International Business Machines Corporation Optimizing dynamical resource allocations based on locality of resources in disaggregated data centers
US10841367B2 (en) 2018-05-17 2020-11-17 International Business Machines Corporation Optimizing dynamical resource allocations for cache-dependent workloads in disaggregated data centers
US10893096B2 (en) 2018-05-17 2021-01-12 International Business Machines Corporation Optimizing dynamical resource allocations using a data heat map in disaggregated data centers
US10936374B2 (en) 2018-05-17 2021-03-02 International Business Machines Corporation Optimizing dynamic resource allocations for memory-dependent workloads in disaggregated data centers
US10977085B2 (en) 2018-05-17 2021-04-13 International Business Machines Corporation Optimizing dynamical resource allocations in disaggregated data centers
US11221886B2 (en) 2018-05-17 2022-01-11 International Business Machines Corporation Optimizing dynamical resource allocations for cache-friendly workloads in disaggregated data centers
US11330042B2 (en) 2018-05-17 2022-05-10 International Business Machines Corporation Optimizing dynamic resource allocations for storage-dependent workloads in disaggregated data centers
CN115442444A (zh) * 2022-11-04 2022-12-06 山东紫菜云数字科技有限公司 一种应用于多对多匹配平台的资源平衡自适应调度方法

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112840608A (zh) 2018-10-22 2021-05-25 康普技术有限责任公司 用于长期演进演进节点b中的分组处理的负载测量和负载均衡

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5881284A (en) * 1995-10-26 1999-03-09 Nec Corporation Method of scheduling a job in a clustered computer system and device therefor
WO2001038972A1 (fr) * 1999-11-22 2001-05-31 Motorola Inc. Procede d'equilibrage de charge dans un reseau de communication
US20070143116A1 (en) * 2005-12-21 2007-06-21 International Business Machines Corporation Load balancing based upon speech processing specific factors
US20070168548A1 (en) * 2006-01-19 2007-07-19 International Business Machines Corporation Method and system for performing multi-cluster application-specific routing
US20090328054A1 (en) * 2008-06-26 2009-12-31 Microsoft Corporation Adapting message delivery assignments with hashing and mapping techniques

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2524550A1 (fr) * 2004-10-26 2006-04-26 Platespin Ltd Systeme pour optimiser l'utilisation des serveurs dans un centre de donnees
US7669029B1 (en) * 2006-11-15 2010-02-23 Network Appliance, Inc. Load balancing a data storage system
JP5359295B2 (ja) * 2009-01-16 2013-12-04 富士通株式会社 負荷分散装置、負荷分散方法および負荷分散プログラム

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5881284A (en) * 1995-10-26 1999-03-09 Nec Corporation Method of scheduling a job in a clustered computer system and device therefor
WO2001038972A1 (fr) * 1999-11-22 2001-05-31 Motorola Inc. Procede d'equilibrage de charge dans un reseau de communication
US20070143116A1 (en) * 2005-12-21 2007-06-21 International Business Machines Corporation Load balancing based upon speech processing specific factors
US20070168548A1 (en) * 2006-01-19 2007-07-19 International Business Machines Corporation Method and system for performing multi-cluster application-specific routing
US20090328054A1 (en) * 2008-06-26 2009-12-31 Microsoft Corporation Adapting message delivery assignments with hashing and mapping techniques

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10159012B2 (en) 2011-11-14 2018-12-18 Alcatel Lucent Baseband signal processing cluster
EP2781116B1 (fr) * 2011-11-14 2019-10-23 Alcatel Lucent Grappe de traitement de signal de bande de base
US10601903B2 (en) 2018-05-17 2020-03-24 International Business Machines Corporation Optimizing dynamical resource allocations based on locality of resources in disaggregated data centers
US10841367B2 (en) 2018-05-17 2020-11-17 International Business Machines Corporation Optimizing dynamical resource allocations for cache-dependent workloads in disaggregated data centers
US10893096B2 (en) 2018-05-17 2021-01-12 International Business Machines Corporation Optimizing dynamical resource allocations using a data heat map in disaggregated data centers
US10936374B2 (en) 2018-05-17 2021-03-02 International Business Machines Corporation Optimizing dynamic resource allocations for memory-dependent workloads in disaggregated data centers
US10977085B2 (en) 2018-05-17 2021-04-13 International Business Machines Corporation Optimizing dynamical resource allocations in disaggregated data centers
US11221886B2 (en) 2018-05-17 2022-01-11 International Business Machines Corporation Optimizing dynamical resource allocations for cache-friendly workloads in disaggregated data centers
US11330042B2 (en) 2018-05-17 2022-05-10 International Business Machines Corporation Optimizing dynamic resource allocations for storage-dependent workloads in disaggregated data centers
CN108710540A (zh) * 2018-05-23 2018-10-26 深信服网络科技(深圳)有限公司 一种分布式集群中的资源调度方法、装置及设备
CN115442444A (zh) * 2022-11-04 2022-12-06 山东紫菜云数字科技有限公司 一种应用于多对多匹配平台的资源平衡自适应调度方法

Also Published As

Publication number Publication date
WO2013178394A1 (fr) 2013-12-05

Similar Documents

Publication Publication Date Title
EP2669798A1 (fr) Répartiteur de charge, gestionnaire de ressources intra-grappes, gestionnaire de ressources intergrappe, appareil de traitement de signaux de bande de base, procédé et programme d'ordinateur pour distribuer une charge
US11438800B2 (en) Apparatus, system and method for traffic data management in wireless communications
US10159012B2 (en) Baseband signal processing cluster
EP3962032B1 (fr) Procédé et appareil d'équilibrage de charge de serveur dans des systèmes ran en nuage
US20180310199A1 (en) Ue load-sensitive front-haul controller
US10887072B2 (en) Methods and apparatus for adjusting a carrier aggregation operation in a wireless communication system
WO2019125255A1 (fr) Procédé et agencement d'assistance d'attribution de faisceau
US10831553B2 (en) System and method for fair resource allocation
Yu et al. Multi-resource allocation in cloud radio access networks
CN109479203B (zh) 控制节点及其方法
Matoussi et al. A user centric virtual network function orchestration for agile 5G cloud-RAN
WO2019080997A1 (fr) Gestion de ressources dans des réseaux d'accès radio en nuage
WO2019081024A1 (fr) Architecture parallèle de planificateur
Koutsopoulos The impact of baseband functional splits on resource allocation in 5G radio access networks
Hirayama et al. RAN slicing in multi-CU/DU architecture for 5G services
KR20130060350A (ko) Atca-기반 장비에서 통신 트래픽을 스케줄링하기 위한 방법 및 장치
Sigwele et al. Call admission control in cloud radio access networks
WO2020187417A1 (fr) Procédés, appareils et programmes informatiques pour la configuration d'un réseau de télécommunication
Wen et al. Protocol stack mapping of software defined protocol for next generation mobile networks
US8320313B1 (en) Method and system for carrier frequency management based on slot contention
CN105101299A (zh) 一种载波聚合系统中负载均衡的方法
Budhdev et al. Isoran: Isolation and scaling for 5g ranvia user-level data plane virtualization
Chabbouh et al. Service scheduling scheme based load balancing for 5G/HetNets Cloud RAN
Gao et al. R-cloud: A cloud framework for enabling radio-as-a-service over a wireless substrate
CN102325376A (zh) 一种载波资源分配方法及装置

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

17P Request for examination filed

Effective date: 20130111

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

AX Request for extension of the european patent

Extension state: BA ME

111Z Information provided on other rights and legal means of execution

Free format text: AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

Effective date: 20140303

RBV Designated contracting states (corrected)

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

RAP1 Party data changed (applicant data changed or rights of an application transferred)

Owner name: ALCATEL LUCENT

D11X Information provided on other rights and legal means of execution (deleted)
17Q First examination report despatched

Effective date: 20160512

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE APPLICATION IS DEEMED TO BE WITHDRAWN

18D Application deemed to be withdrawn

Effective date: 20160923